Edit report at https://bugs.php.net/bug.php?id=61182&edit=1
ID: 61182
Comment by: phpmpan at mpan dot pl
Reported by: tom at tomwardrop dot com
Summary: Assume Opening PHP Tag
Status: Open
Type: Feature/Change Request
Package: *Configuration Issues
PHP Version: 5.4.0RC8
Block user comment: N
Private report: N
New Comment:
Note: I'm NOT against the idea itself. I'm just thinking that in the current form it can
do more harm than good.
What you're asking for is redefining the whole PHP world. Let's imagine that PHP6 includes
your idea and it's *not* optional. What happens?
1. Many of PHP books become obsolete
We all know mixing code and output is bad, but the books
take this approach because it's simpler. It allows authors
to show the basic ideas of PHP without requiring the reader
to download/install third party template engine.
But if .php files are no longer templates, books need
to be rewritten. Lots of money for authors, but I think
it's not dev's goal.
2. Lots of currently used code becomes obsolete
If one needs to write code for a server that has
this feature enabled, any template-like code should
be avoided. This means we can use only "safe" libraries.
Which are "safe"? Only those for which the author states
they're compatible with assume_open_tags.
In other words: less code for us to use. Many things needs
to be rewritten. This is bad.
3. Admins will simply refuse to enable the feature
I love the idea of removing magic_quotes. At the same time
I believe many admins will hesitate to upgrade to PHP6,
because they have irrational belief that the magic_quotes
feature was protecting them.
Now imagine what will they do with assume_open_tags.
Will they enable it? Will they risk breaking already
deployed applications? I don't think so. If they're afraid
to leave their servers without magic_quote "protection",
they'll be even more scared of the fact that they can beak
something seriously by enabling assume_open_tags.
Setting assume_open_tags on per-directory basis (for example
with .htaccess in Apache) doesn't solve the problem, because
PHP libraries may be shared between multiple applications.
I believe that books should be rewritten, real template engines should be used, we should update our
code et cetera. But real life is real life. Encountering pieces of software that were not upgraded
for 20-30 years is not an uncommon thing ("20-30yrs" does not apply to PHP, but I know
apps that were not updated since PHP4). magic_quotes are deprecated for years and many
people seen they're bad even earlier. There was enough time to update applications that depends
on them. And even if some code is not fixed, removing magic_quotes doesn't make it stop
working. The case of assume_open_tags is different. If it's optional, it needs to
become a standard to be accepted. And this should be done quickly. I can't imagine building
separate versions of libraries for server with this feature enabled and without it. Authors will
simply keep using versions with "<?php" to maintain compatibility and the proposed
feature will stay unused.
OTOH forcing it to be enabled will cause problems mentioned above. This is a lose-lose situation.
IMHO this may work only if the author of the code decides which mode to use. This makes the feature
really optional. It may be included in the server without breaking any existing code and be enabled
if new code requies it. This way the feature may be introduced gradually. Evolution, not revolution.
The question is: how to enable authors to tell that their code assumes that opening tags are open?
The first idea was to add some metainfo to a PHAR.
The second, based on your last post, is to add an optional argument to include*/require* constructs.
In such case the following code would cause included file to be parsed as a raw PHP source, not
requiring additional "<?php".
require_once some_magic 'ns1/ns2/Example.class.php'
There still needs to be a file with "<?php" at the begining to use this code. However
currently the trend is to use a single dispatcher, so it's not a big deal.
Still I'm not sure if the feature is really worth being implemented.
Previous Comments:
------------------------------------------------------------------------
[2012-02-26 00:12:45] tom at tomwardrop dot com
It's not just about the extra characters, but like the end ?> tag (which
thankfully is optional), any white-space or otherwise non-printable characters
before the opening tag can cause "headers sent" issues. You could solve that
problem by implementing the ignore white-space rule I've already mentioned,
where any white-space before the opening tag is ignored.
The more I think about this and talk to the others, the more it becomes apparent
that what I'm actually asking for, is a distinction to made between PHP
templates, and PHP scripts/applications. If PHP were to define these two
distinct concepts, then you could do more than just make the opening tag
optional. For example, you could have a template() or render() method to act as
an include() for php templates. Unlike include() however, this render() method
would return the output of the file, instead of sending it straight to the
browser. This would negate the need to capture template output using the output
buffer functions (something that I believe most frameworks end up using).
Making such a distinction would also allow web servers like Apache to treat PHP
files differently. You may create a rule in Apache to render all .phpt files as
PHP templates, rendering everything else as PHP script or application files. We
may then see mod_php implement an application mode, where one can define a
single-point of entry into their application. This could have flow-on
performance benefits, where mod_php could cache the parsed PHP, then either fork
it on each request, or instantiate a new application class. Such a feature would
mean frameworks wouldn't have to stuff around with .htaccess files, and would
mean that programmers don't need to add the following to the top of all their
files: if (defined('SOME_CONSTANT')) exit;
While there's momentum among the PHP developers to move forward with modernising
the language, I think now would be a good idea to consider some of these more
fundamental changes. PHP's built-in template engine, ease of deployment, and
it's dynamic, traditional OO constructs would still remain PHP's strengths.
With all this said, I'd be happy to save such changes to a major release
intended to break legacy code, like PHP 6. I'd like to keep in mind too that
code portability isn't relevant to most people who don't intend to release their
code as open source. Typically, those people using PHP in a business context
have control of their server. It's only shared hosting environments where
portability becomes a potential issue. All I'm saying is don't rule out ideas
based on the lowest common denominator.
------------------------------------------------------------------------
[2012-02-25 14:12:03] phpmpan at mpan dot pl
After a bit of thinking I came to a conclusion that PHARs can, in theory, have such thing
implemented. Some metadata may be included in PHAR to tell PHP that every source file in the archive
assumes "<?php" tag to be open. Since such information is included by the author of the
code, nothing can be broken.
However I don't know if it's worth being implemented. As I said before, it gives almost
nothing. Five characters is not much if one have to put dozen lines at the begining of each file.
Also PHARs, that assume opening tag to be open, should be incompatible with older versions of PHP to
prevent sending source code to the client by accident. Too much trouble IMO.
------------------------------------------------------------------------
[2012-02-25 13:58:34] phpmpan at mpan dot pl
Sorry, I have misunderstood the "<?php>" tag. Thought it's used to *enable*
assume_open_tag. But, if you want to use it to disable the feature, it's even
worse. This breaks lots of existing code. Mixing PHP with HTML is an example of bad design, but
there are lots of such things and PHP devs can't just say "from today your code is not
working, because we have decided to break it". Adding some magical sequence of characters at
the begining is not a solution for this problem. Do you imagine administrators going through all of
the files on their servers and adding "<?php>" to fix the broken code? Even harder
to imagine in case of obfuscated or PHARed code.
I believe there is enough problems with server incompatibilities already. No need to make next one.
I would be much more happy to see UTF-8 to be a standard feature on every host. A requirement to
write "<?php" doesn't bother me. Copyright notices and namespace
declarations/imports use much more space.
------------------------------------------------------------------------
[2012-02-25 13:44:12] phpmpan at mpan dot pl
So, basically, you're suggesting that programmers should write "<?php>" at the
begining of the file to not write "<?php". Blarg.
Of course this "<?php>" thing is optional. The problem is that virtually any code
has to use it to be portable. This means it's not really optional.
------------------------------------------------------------------------
[2012-02-25 09:24:21] tom at tomwardrop dot com
Are there not other directives that can break a lot of code? Remember, this
would default to off. I don't see why as a server owner, I should have this
option made unavailable purely because it can break other code. If you wanted to
write code that worked regardless of this setting, you could do something like:
<?php
init_set('implicit_open_tag', false)
?>
Of course, for that to work then implicit_open_tag is On, the parser would have
to ignore the "<?php". The rule could be that if <?php is the first non-
whitespace sequence encountered in the file, then it's ignored.
'optional_open_tag' may therefore be a more appropriate name for this setting.
Except for legacy templates which may start with something other than <?php,
this would allow for cross-environment code. Any such template code that breaks,
would break in a manner no different to how new features like namespaces break
in older version of PHP.
A new tag could be introduced: "<?php>". This would be shorthand for opening and
closing a php tag, and should be placed at the top of any template file that has
the requirement to work regardless of whether the opening tag is optional.
I hope this idea isn't dismissed on the grounds that it's difficult to
implement, because I think it's workable. Having optional opening tags would no
doubt be a step in the right direction for PHP, and I'm sure that if you didn't
have backwards compatible to be concerned about, you'd probably make opening
tags implicit with no option to make it otherwise. As I said earlier, the
decision to make the opening tag explicit was desirable at the time PHP was conceived, but I believe
it's one of those legacy decisions that needs to be re-
evaluated.
------------------------------------------------------------------------
The remainder of the comments for this report are too long. To view
the rest of the comments, please view the bug report online at
https://bugs.php.net/bug.php?id=61182
--
Edit this bug report at https://bugs.php.net/bug.php?id=61182&edit=1